This paper from a perspective of fine-grained sentiment,aims to survey about the abstraction of product attributes and emotional vocabulary,the sentiment analysis at chapter and sentence levels,the building of corpus and sentiment lexicon and the machine learning.Then we point out that the efforts to build corpus and sentiment lexicon in China is already under way.However,experts participation in building basic corpus for products of different industries is still needed,considering the great varieties of products and the underperformed portability of core attributes.The achievements of NLP algorithm should be widely used in the fine-grained sentiment analysis and the algorithms portability should be greatly improved.The fine-grained sentiment analysis on online customers reviews should take marketing research topics as its core,and expand and innovate the marketing theories so as to avoid blind pursuing of the accuracy and complexity of algorithm.